> After increasing the number of CPUs in the virtual machine the failure no longer reproduced. It is reproducible only with a single CPU.

The test just failed on AMD64 Debian root 3.x and according to test.pythoninfo, this machine has a single CPU:

os.cpu_count: 1

http://buildbot.python.org/all/#/builders/27/builds/1247

...
== CPU count: 1
== encodings: locale=UTF-8, FS=utf-8
Using random seed 6423496
Run tests in parallel using 2 child processes
0:00:00 load avg: 0.71 [  1/417] test_genericclass passed
0:00:00 load avg: 0.71 [  2/417] test_copyreg passed
0:00:03 load avg: 0.71 [  3/417] test_os passed
stty: standard input: Inappropriate ioctl for device
running: test_tokenize (32 sec 627 ms), test_signal (30 sec 11 ms)
0:00:51 load avg: 1.32 [  4/417] test_tokenize passed (50 sec 821 ms) -- running: test_signal (48 sec 506 ms)
0:00:54 load avg: 1.32 [  5/417/1] test_signal failed
...
test_stress_delivery_dependent (test.test_signal.StressTest) ... detected median itimer() resolution: 0.000044 s.
FAIL
test_stress_delivery_simultaneous (test.test_signal.StressTest) ... detected median itimer() resolution: 0.000032 s.
ok
...
======================================================================
FAIL: test_stress_delivery_dependent (test.test_signal.StressTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/root/buildarea/3.x.angelico-debian-amd64/build/Lib/test/test_signal.py", line 1156, in test_stress_delivery_dependent
    self.assertEqual(len(sigs), N, "Some signals were lost")
AssertionError: 3617 != 10000 : Some signals were lost